Generally, in coding processing of a moving picture, the amount of information is reduced by compression for which redundancy of a moving picture in spatial direction and temporal direction is made use of. Generally, conversion to a frequency domain is performed as a method in which redundancy in spatial direction is made use of and coding using prediction between pictures (the prediction is hereinafter referred to as inter prediction) is performed as a method of compression for which redundancy in temporal direction is made use of. In the inter prediction coding, a current picture is coded using, as a reference picture, a coded picture which precedes or follows the current picture in order of display time. Subsequently, a motion vector is derived by performing motion estimation on the current picture with reference to the reference picture. Then, redundancy in temporal direction is removed using a calculated difference between picture data of the current picture and prediction picture data which is obtained by motion compensation based on the derived motion vector (see Non-patent Literature 1, for example). Here, in the motion estimation, difference values between current blocks in the current picture and blocks in the reference picture are calculated, and a block having the smallest difference value in the reference picture is determined as a reference block. Then, a motion vector is estimated from the current block and the reference block.